Regarding your last paragraph (And as long as...) Bill Gates convincingly argues that the traits that make AI do good things are not easily separable from those that make it do bad things. So it may not be possible to explicitly avoid an AI that does bad things.